Levenshtein distance

Let’s take a deeper look at the Levenshtein distance between “nyuiin” original onomatopoeia and the new “nyuiin” re-onomatopoeia. The Levenshtein distance is a string metric for measuring the number of changes (such as swaps, deletions, and additions of characters) needed to change one word into another. In this case, we examined our data using an index where 1 represents total similarity and 0 represents no similarity.

You can see the results below, where the Levenshtein distance method did not detect much similarity. However, II think we might get an entirely different picture if we measured the degree of similarity using another metric.

Chuku-Chuku – 0.000000
Nyuuu – 0.000000
Nyoki-Nyoki – 0.333333
Myooh – 0.400000
Funyai – 0.200000
Kyumofu – 0.000000
Powaan – 0.200000
Monyopo – 0.000000

Correlation heatmap

These heatmaps graph out the degree of correlation among different attribute groups such as “ofyuchi”, “barimetaa”, and “nyuiin”. To understand how to look at the data, take for example the topmost row, which allows you to see hardness and its correlation to other attributes. Using the topmost row as our example, then within a Nyuuin environment, hardness and viscosity are positively correlated, so we might imagine that a Nyuuin texture is both firm and sticky.

Gestalt pattern matching

This is a method for determining the degree of matching between two strings by taking the longest common substring and then recursively using the same process on the strings leftover before and after the substring.

If these don’t have the same repetitions as the original, they won’t be considered a match, so this method doesn’t usually result in complete matches, but “nyuiin” had a relatively high number of matches.

Nyuiin
——–
0.0
0.0
0.36363636363636365
0.5
0.4444444444444444
0.0
0.2222222222222222
——–
Ofyuchi
——–
0.0
0.0
0.0
0.0
0.0
0.0
0.0
——–
Barimetaa
——–
0.0
0.0
0.0
0.0
0.2222222222222222
0.0

Jaro-Winkler distance method

This method is a little bit similar to the Levenshtein distance method, but calculates the distance based on the total number of characters in common as well as characters which don’t need to be transposed. Because of this, we can say that the partial matches for our example strings are significant.

Nyuiin
——–
0.0
0.0
0.6622222222222223
0.6888888888888888
0.6333333333333333
0.0
0.48333333333333334
——–
Ofyuchi
——–
0.0
0.0
0.0
0.0
0.0
0.0
0.0
——–
Barimetaa
——–
0.0
0.0
0.0
0.0
0.48333333333333334
0.0

Levenshtein method

To be honest, we probably shouldn’t get too caught up with the Levenshtein method. It takes into account the number of swaps and deletions, so it is probably the best for capturing nuance, but I also feel it may be a bit too complicated to explain and use.

Ofyuchi
——
Kyuwan – 0.0
Moyon-Moyon – 0.0
Mochi-Mochi – 0.0
Fuwapu – 0.0
Nuchofunya – 0.0
Noko-Noko – 0.0
Kyui – 0.0
BuchuReʔʔ – 0.0
Barimetaa – 0.0
——
Pyuyon – 0.0
Gyu-Gyu – 0.0
Boro-Boro – 0.0
Nyomo-Nyomo – 0.0
Goririn – 0.2
Mochohonii – 0.0

These three methods of parsing strings have all developed alongside search engines, so they have a much stronger application at correcting subtle omissions and misspellings. For this reason, it can be argued that it might be a bit of a tall order to use them in a case like ours, where we want to find matches between words with similar nuances.

Despite this, our efforts here showed surprising results in the case of “nyuiin” and its high matchrate.

In particular, when using the Jaro-Winkler method, we derived an approximately 0.69 rate of similarity with the “myooh” onomatopoeia. The nuance isn’t really conveyed by language, but isn’t it interesting to get a similarity rate as high as this?

Using the numerical data for the six texture attributes of oden, there’s a possibility we could present some new onomatopoeia based on the correlation heatmap.

Let’s take once more the “nyuiin” onomatopoeia as an example. First, we have discovered that the “nyuiin” oden shape features a positive correlation between hardness and viscosity.

This means that we can predict its 3D structure will be both firm and sticky. Therefore, we should be able to use this data to reverse-engineer onomatopoeia that will correlate to firm, sticky textures. By separating the attributes numerically, it is possible in theory to create onomatopoeia which are even more suited to the printed structures.

This system may be applied within companies for copyright and product name development (although I can’t say if they would be willing to invest so much effort into it).

Anyway, let’s call our reverse-engineered onomatopoeia “re-re-onomatopoeia”, shall we?
All right then, so it might be a bit more interesting to look at some examples that leave out the values from our initial onomatopoeia.

Using the language processing logic, we went ahead and used three methods to measure the correlation between reverse-engineered onomatopoeia and the human-generated onomatopoeia.

Unfortunately, we didn’t find any high rate of similarity between them, but we did get interesting results for some of the onomatopoeia. When using the Jaro-Winkler method for “nyoiin,” there were four onomatopoeia out of seven that showed slightly high similarity rates. One of the indicators seemed to point towards the existence of a “buubakiki” sort of onomatopoeia that might possibly have a common correlation to food texture sound effects in general. The use of prolonged vowels and “na, ni, nu, ne, no” sounds, or even the shape of the written word itself might also be relevant factors to consider.

Born from the Experimental meetings of the 3D Structure Oden Society held between January and March of 2022, the Dawn of Computational Food Texture Design is a development collective that combines digital fabrication technology and computational design to create new textural and food creation processes.

We’re trying to create a world in which edible onomatopoeia are commonplace, by facilitating interdisciplinary discussion and prototyping among individuals well-versed in different areas of expertise such as haptics, computational design, natural language processing, machine learning and digital fabrication.

Want to get in touch with the Metamaterial Design Technology Research Team, Melt.?

Project participation inquiries:

You can apply to participate in our project through our Facebook Group.
Please feel free to browse around event announcement and activity report posts

For inquiries relating to our various projects and to collaborate with Melt:

At Melt. we call “X Projects” all the initiatives and events we organize with the cooperation of outside groups. This is an intuition-driven framework which includes a wide variety of activities in many different formats of metamaterial design, and everything from research to prototyping. If you want to get out there and search for new possibilities without being tied to old patterns of thought, if you’re looking get away from desk-bound theory and get your hands on some actual practice with real projects that close the loop between planning and implementation… By all means, please go ahead and contact us via the form below.

Melt. is also promoting design with materials that capture physical phenomena, so we are always on the lookout for measurement equipment, databases, and other resources we can make use of.

We would love to hear from companies willing to lend us point-cloud data, 3D scanners, microphones for collecting sounds, processing software and more. If you happen to work for such a company, and think that Melt.’s activities look like fun, or that you’d like to join in our discussions, not just provide equipment, please reach out to us at the email address below.

What Melt. can do for you…

• We use technology to create undiscovered design methods that can be applied to exploring new sales channels.
• Through our conferences and workshops, we help grow and nurture user communities in a wide range of topics.
• “I didn’t know you could use it like that!” Through our projects, we generate bursts of insight that lead to all kinds of unsuspected methods and applications.

Contact

info.nagoya@fabcafe.com（Saito）